- Added a method to send failure notifications to Flutter when the background service fails to enter the foreground.
- Implemented a broadcast receiver to handle service failure notifications and notify Flutter about the failure.
- Enhanced the persistent streaming service to attempt recovery for failed streams.
- Introduced heartbeat monitoring for SSE streams to detect stale connections and trigger recovery actions.
- Added a method to ignore specific errors related to RenderFlex overflow.
- Refactored the error display layout to use a centered scrollable view.
- Improved visual presentation with a gradient background for the error icon and updated button styles for retry functionality.
Introduce scrimMedium and scrimStrong color tokens to the shared
theme color tokens and propagate them through constructors, copyWith,
and lerp so scrim values interpolate and can be overridden.
Define scrim tokens as black with different alpha values per theme
mode (lighter alpha in light mode stronger in dark mode) to create
a consistent darkening effect for overlays.
Refactor ChatPage to use the new scrim tokens for drawer scrims and
format a RegExp call for readability. This replaces previous use of
overlay tokens for platform-specific scrims to provide clearer
semantics and better visual control for modal/drawer backdrops.
the modern chat input composer layout to better handle vertical
space and alignment. Change the row crossAxisAlignment from
center to end so controls align to the input bottom. Wrap the animated
container with a ConstrainedBox that limits the composer max height to
25% of the screen, preventing excessive growth on tall displays.
Reorder and clean up the AnimatedContainer/decoration block, extracting
the composer radius constant and preserving styling (background color,
border color/width). Move the text field and inline mic into the same
row structure under the constrained container and retain the isActive
flag and padding. These changes prevent layout overflow, ensure a
consistent border radius, and improve visual alignment of controls.
Remove internal reasoning from copied message text to avoidleaking implementation details or developer-only when a
user content from the chat- In chat_pagecopyMessage, cleaning steps to:
- <details type="ing">...</details blocks
- <think>think> and <ing>...</reason> tags
- trim leftover whitespace before writing to the clipboard
- In assistant_message_widget._buildSegmentedContent, remove an
unused hasMediaAbove calculation and a conditional spacer that added
extra top padding before reasoning tiles. This simplifies rendering
logic and avoids relying on removed spacing behavior.
Hide archived assistant variants in the linear chat view and track
previous assistant as versions so regenerated responses do not
duplicate or lose history. When regenerating, mark the previous assistant
message with an archivedVariant flag for the UI and keep it in server
history. Add a ChatMessageVersion model and a versions field to
ChatMessage to store prior generated variants. Implement
archiveLastAssistantAsVersion in chat providers to snapshot the last
assistant message into versions and reset the message for a fresh
streamed generation. Finalize flow updates to attach an adjacent archived
assistant as a version when needed so the UI can present a switcher
between current and past variants. These changes prevent duplicate
messages, preserve previous responses, and enable variant switching.
Add Riverpod-aware image header and cache manager support for network
images used in avatar and markdown widgets. Convert AvatarImage to a
ConsumerWidget and read a self-signed cache manager and HTTP headers
from Riverpod so CachedNetworkImage can send auth/custom headers and use
the provided cache manager. In Markdown image builder, obtain headers
and cache manager from a ProviderContainer (via ProviderScope.containerOf)
to enable the same authenticated loading in non-consumer contexts.
Introduce image_header_utils.dart to centralize building Authorization
and custom headers from auth/api providers, with helpers for Ref,
WidgetRef, and ProviderContainer. Add dependency adjustments in
pubspec.lock for flutter_cache_manager and http marked as direct main.
These changes ensure protected images (self-signed or auth-required)
load correctly across the app and reuse the configured cache manager.
Add rendering and support for a karaoke-style text-to-speechprogress bar in assistant messages so users can see the currently
spoken sentence and highlighted word during playback.
- Append TTS karaoke bar to AssistantMessageWidget when the message is
the active TTS target and playback is speaking/paused/loading.
- Implement _buildKaraokeBar to render the active sentence with a
highlighted word span, using ConduitCard and theme styles.
- Import conduit_components for shared UI primitives.
- Extend TextToSpeechState with sentence data:
sentences, sentenceOffsets, activeSentenceIndex, and per-word
progress (wordStartInSentence, wordEndInSentence).
- Add provider callbacks wiring: onSentenceIndex and
onDeviceWordProgress handlers (hooked into TTS backend).
- Prepare sentence splitting and word-progress plumbing in the TTS
provider (prepares data used to drive the karaoke display).
This change improves UX by visually indicating the spoken sentence
and current word during TTS playback, aiding comprehension and
accessibility.
queued playback via a dedicated _startServerChunkedPlayback method
instead of generating a single monolithic audio blob.
This change simplifies the server flow, avoids constructing an entire
audio buffer in memory, and enables smoother playback and error
recovery. On errors, the code still falls back to device TTS.
